Auto-adapting pest deterrent system using artificial intelligence

The invention relates to location monitoring generally and, more particularly, to a method and/or apparatus for implementing an auto-adapting pest deterrent system using artificial intelligence.

Pests can ruin a potentially great experience. From cottages that get swarmed by bugs at particular times of year to decks that get covered in the droppings from various animals, pests are undesirable. Aside from being a nuisance, pests can pose a danger to particular activities, such as birds and airports. Pests can also cause economic harm, such as eating valuable crops.

While it is desirable to remove pests, important considerations should be involved. The harm should not be worse than the cure. Pesticides can harm humans. Objects like scarecrows can be unsightly. Loud noises used to scare birds can be a nuisance. Efforts should also be made not to harm the pests. Simply because a pest is annoying, does not mean that the pest should be killed or harmed. Harming pests can also have unintended consequences (i.e., pesticides meant to protect crops also harm crops by reducing the population of pollinators).

Many animals are capable of adapting to particular deterrents. An owl statue might be effective initially, but pests will eventually not be afraid of the owl statue. Crows are notoriously intelligent and can adapt to a deterrent. If the deterrent methods are not safe or stop deterring the pests, then the deterrent is not effective.

It would be desirable to implement an auto-adapting pest deterrent system using artificial intelligence.

The invention concerns an apparatus comprising an interface and a processor. The interface may be configured to receive sensor data from a plurality of sensors. The processor may be configured to detect an intruder in response to an analysis of the sensor data, activate an AI model in response to detecting the intruder and generate a countermeasure signal in response to a countermeasure selection by the AI model. The AI model may be configured to analyze the sensor data, compare the sensor data of the intruder to a database of pests to perform a classification of the intruder, determine the countermeasure selection in response to the classification of the intruder as a selected pest, monitor the sensor data for an outcome of the countermeasure selection, and generate a text description of the classification of the selected pest, the countermeasure selection and the countermeasure outcome.

Embodiments of the present invention include providing an auto-adapting pest deterrent system using artificial intelligence that may (i) detect a presence of a potential pest, (ii) classify the potential pest against a list of known pests, (iii) deploy a particular deterrent in response to a classified pest, (iv) deploy deterrents that are non-lethal, (v) adapt deterrence methods to particular pests over time through learning, (vi) record events comprising a pest detection, a deterrence method used and an effectiveness of the deterrence method, (vii) analyze a database of deterrence events recorded using artificial intelligence (AI), and/or (viii) be implemented as one or more integrated circuits.

Embodiments of the present invention may be configured to use a combination of sensor inputs and artificial intelligence (AI) models to detect the presence of pests and/or other nuisances and deploy active, non-lethal countermeasures to encourage the pests to leave. Embodiments of the present invention may be implemented in a pre-defined area (e.g., a geo-fenced area). In one example, a detected pest may be geese that enter a water front of a home. In another example, a detected pest may be swarms of bugs in an outdoor region. In yet another example, a detected pest may be a skunk on a property. The types of pests detected may be varied according to the design criteria of a particular implementation.

The sensors implemented may be configured to detect a potential pest. The AI model may be configured to classify the potential pest. For example, a database may be implemented comprising a list of known pests and/or effective countermeasures for particular pests. The AI model may access the database to determine the particular pest detected. If a particular type of pest is detected, then the AI model may select an effective countermeasure to encourage the invader to leave the area. The countermeasures may be non-lethal. The countermeasures may be a series of countermeasures that may escalate over time. The countermeasures may be selected to be as non-intrusive to people as possible. In some embodiments, the deterrents may be contained in fixed devices strategically placed on the property. In some embodiments, the deterrents may be implemented as mobile platforms (e.g., automated air drones, automated ground bots, fixed installation deterrents, etc.) that may be limited to the geo-fenced space. In one example, the deterrent selected may be audio that may be in a frequency that is inaudible to humans. In another example, the deterrent selected may be lights. In yet another example, the deterrent selected may be emitting a smell. In still another example, the deterrent selected may be deploying a device capable of aggressive posturing. The type of deterrent(s) selected may be varied according to the design criteria of a particular implementation.

Embodiments of the present invention may be configured to log (e.g., record) the events detected. Each event record may store the sensor data of the event, the pest classification, the countermeasure deployed and/or an effectiveness of the countermeasure selected. The data for the event records may be provided as input to an AI model (e.g., a Large Language Model (LLM)). The LLM may be configured to generate a natural language text description of the event and the outcome of the event. The natural language text description may comprise a human readable description of the event with sufficient detail to enable a visually impaired person to understand the contents of the event data (e.g., using a screen reader). For example, the sensor data (e.g., video data, lidar data, thermal image data, radar data, etc.) may be converted to readable text.

Embodiments of the present invention may be configured to gather the database the events in the database (e.g., comprising the natural language description of the events) and provide the events as input to an artificial intelligence model (e.g., an LLM) to analyze the events. In some embodiments, the LLM may be the same LLM implemented to generate the natural text description of the events. In some embodiments, the LLM may be a separate AI model from the LLM used to generate the natural text description of the events. The event records may be provided to the LLM continuously (e.g., in real-time) or periodically. The LLM may analyze the events, collect common events, evaluate the effectiveness of the deterrent(s) applied and/or learn about the effectiveness of the deterrent(s) used over time.

Based on the analysis of the aggregation of the events and/or the effectiveness of the deterrents over time, the LLM may learn how the particular pests have adapted to the currently used deterrence methods and manage the recommended deployment of the deterrence methods. In one example, the recommended deployment of the deterrence methods may comprise rotating the deterrent selected (e.g., to avoid continuously re-using a single deterrent method). In another example, the recommended deployment of the deterrence methods may comprise narrowing down options to reach an estimated true deterrence method (e.g., optimize for future engagements). Generally, the recommended deployment may comprise determining how to handle future engagements in similar circumstances such that the efficiency and efficacy of the system improves over time to match the deterrent employed with the pest detected (e.g., learn such that as pests evolve to one approach or the efficacy of an approach declines, the system may evolve as well to continue to be effective over time).

Embodiments of the present invention may be configured to determine whether an intruder is a pest. For example, a desired guest or a “friendly” person (or animal) may be detected. The deterrents may be deployed when the intruder is classified as a pest. In some embodiments, the selection of the particular deterrent deployed may be affected by the presence of a desired guest. For example, particular deterrent methods may not be deployed if people are also present (e.g., a smell may not be deployed when people are in the area, sprinkler systems may not be activated when people are in the area, etc.). In another example, the time of day and/or time of year may affect the selection of a particular deterrence methods (e.g., sprinkler systems would not be used in the winter, lights may be more useful at night, etc.). Generally, the non-lethal deterrents ideally may leverage sounds, light and/or other sensory stimulus on the pests that are out of the human range of human sensory perception (e.g., sound frequencies that animals can hear but are outside of the human hearing range, such that the system does annoy humans in the process of deterring pests).

Embodiments of the present invention may be configured to store the events. The storage and/or observation process may be automated. The event records may define an event as successful and/or unsuccessful. The event records may measure an amount of time taken to encourage pests to leave and/or measure other parameters that may be monitored over time. The efficacy and/or efficiency of the deterrent method may be monitored. The LLM may monitor the event logs over time to learn how efficacy to particular deterrents by particular pests change over time. If the efficacy of a deterrent against a pest degrades over time, the system may automatically adapt and modify the deterrent used in future engagements (e.g., on an as-needed basis). The process may repeat over time to continually adapt and/or evolve.

Referring to, a diagram illustrating an example embodiment of the present invention is shown. An environmentis shown. The environmentmay provide an illustrative example of a dock area. The environmentmay comprise a sky region, a water regionand/or a dock. A boatmay be moored to the dock. The environmentmay be a representative example of an area that may be used by people that may be affected by pests. For example, people may use the dockto access and launch the boatonto the water region. In the example shown, no people may be currently present in the environment.

A number of intruders-are shown in the environment. In the example shown, the intruders-may be geese. The geese-are shown on the dock. Geese droppings-are shown on the dock. The geese droppings-may affect the usability of the dock. For example, the geese droppings-may be unsightly, may be disgusting and/or may potentially spread disease. The geese-may be examples of pests. Generally, people may desire to have the geese-removed and/or encouraged to avoid landing on the dock(e.g., to prevent the geese droppings-).

A block (or circuit), blocks (or circuits)-and/or devices (or circuits)-are shown. The circuitmay implement an apparatus, device and/or system. The circuits-may implement sensors. Devices-may implement deterrent devices (e.g., countermeasure devices). Lines-are shown extending from the sensors-, respectively. The lines-may represent a detection region/range of the respective sensors-. In the example shown, the apparatusand the sensors-may be implemented on the boat. In the example shown, the deterrent devices-are shown flying in the sky region. The location and/or implementation of the circuits-may be varied according to the design criteria of a particular implementation.

The apparatusmay be configured to implement an auto-adapting pest deterrent system using artificial intelligence. The apparatusmay operate in a pre-defined area. The pre-defined area may be a geo-fenced region. In the example shown, the environmentmay be an example of a geo-fenced area. The sensors-are shown capturing sensor data of the environment. Data generated by the sensors-(e.g., sensor data) may be presented to the apparatus. The apparatusmay be configured to detect events in response to an analysis of the sensor data. The events may comprise a detection of a potential intruder.

The apparatusmay be configured to perform a classification of events in response to analyzing the sensor data generated by the sensors-. The classification may determine whether the potential intruders detected may be false positive events, desired people/animals, pests, etc. In the example shown, the geese-may be classified as pests. While the geese-are shown as a representative example of intruders and/or pests, the type of intruders and/or pests detected may be varied according to the design criteria of a particular implementation.

The sensors-are shown mounted on the boat. The sensors-may be installed at and/or around the environmentto enable the sensor ranges-to detect potential intruders and/or an event. In one example, the sensors-may be configured to provide raw sensor data to the apparatusand the apparatusmay analyze the sensor data to detect potential intruders and/or an event. In another example, the sensors-may be configured to detect threshold conditions (e.g., to trip the sensor) and in response to detecting the threshold conditions, the sensors-may activate the circuitand/or other sensor systems (e.g., wake from a low power and/or sleep mode of operation). In the example shown, two of the sensors-may be implemented to monitor the environment. In some embodiments, more or fewer of the sensors-may be implemented. The number of the sensors-implemented may be varied according to the design criteria of a particular implementation.

In response to classifying the potential intruders of the event as a pest, the apparatusmay be configured to deploy the deterrent devices-. The deterrent devices-may be implemented to attempt to encourage the pests-to leave the environment. The deterrent devices-may be configured to implement one or more non-lethal deterrent actions. In the example shown, the deterrent devices-may be drones flying in the sky region. For example, the drones-may fly close to the geese-to chase the geese-out of the environment. Other deterrent actions may be implemented. For example, the drones-may be configured to emit noises to chase the geese-away. In another example, the drones-may comprise a consumable (e.g., water). For example, the drones-may be configured to spray water at the geese-. The drones-may be further configured to spray the water at the droppings-to clean the deck(e.g., the deterrence may allow the pests to stay in the environmentbut remove a particular type of behavior and/or an undesirable result of a particular behavior such as the droppings-). While two of the drones-are shown as a representative example of the deterrent devices, other types of deterrent devices and/or more or fewer deterrent devices may be implemented. The number and/or type of deterrent devices implemented may be varied according to the design criteria of a particular implementation.

Referring to, a block diagram illustrating an example embodiment of the present invention is shown. The geo-fenced regionis shown. A block diagram of the apparatus (or system)is shown implemented in the geo-fenced region. For example, the apparatusmay be configured to monitor for and deter pests in the geo-fenced region.

The apparatusmay comprise the sensors-, the deterrence device, blocks (or circuits)-, a block (or circuit), a block (or circuit)and/or a block (or circuit). The circuits-may implement an interface. The circuitmay implement a processor (or CPU, or APU, or GPU, or AI accelerator, etc.). The circuitmay implement a memory. The circuitmay implement a communication device. The apparatusmay comprise other components (not shown). The number, type and/or arrangement of the components of the apparatusmay be varied according to the design criteria of a particular implementation.

The sensors-may be configured to capture sensor data of the environment. The sensors-may be configured to generate a respective signal (e.g., SD_A-SD_N). Each of the signals SD_A-SD_N may comprise the sensor data generated by a respective one of the sensors-. The sensors-may be configured to communicate the sensor data SD_A-SD_N to the interface-. In some embodiments, the sensor data SD_A-SD_N may comprise raw data. In some embodiments, one or more of the sensors-may be edge devices capable of performing analysis of the raw sensor data, and the sensor data SD_A-SD_N may comprise an interpretation of the environmentby the sensors-

The sensors-may comprise various types of sensors and/or may be configured to capture various types of sensor data. In some embodiments, each of the sensors-may be one type of sensor (e.g., all cameras). In some embodiments, one or more of the sensors-may implement different types of sensors (e.g., one or more cameras, one or more motion detectors, one or more lidar devices, etc.). The sensor data SD_A-SD_N generated by the sensors-may provide information about the environmentthat may be used by the apparatusto detect events, people, animals, intruders, audio, etc. In some embodiments, one of the sensors-alone may not be sufficient to detect an event. However, the processormay be configured to implement sensor fusion that may enable inferences to be made based on a combination of data types generated by the sensors-

The sensors-may comprise one or more of a camera device, a radar device, a lidar device, a thermal camera device, a time of flight device, etc. In some embodiments, one or more of the sensors-may provide a threshold event detector. The threshold event may enable one or more of the sensors-to wake up one or more of the components of the apparatus. For example, one of the sensors-implementing a radar device may be operational while the rest of components of the apparatusare in a low-power standby mode of operation, and the radar device, in response to a detection, may be configured to activate the rest of the sensors-and/or one or more components of the apparatus. The number of the sensors-implemented and/or the types of sensors-implemented may be varied according to the design criteria of a particular implementation.

The interface-may be configured to receive the sensor data SD_A-SD_N from the sensors-. The interface-may comprise various inputs/outputs of the apparatus. In an example, the interface-may comprise pins, wires, connectors, etc. The interface-may be configured to generate a signal (e.g., IN). The signal IN may comprise the data received from the sensors-. The signal IN may be generated in response to the signals SD_A-SD_N. The signal IN may be presented to the processor.

The processormay be configured to receive data, perform operations on the data and/or generate output. The processormay receive input from and/or generate output for the interface-, the memoryand/or the communication device. The processormay be implemented as a central processing device (CPU) and/or an accelerated processing unit (APU). The processormay comprise hardware configured to provide hardware-accelerated artificial intelligence (AI) operations.

The processormay be configured to detect an intruder in response to an analysis of the sensor data received in the signal IN from the interface-. In response to the analysis of the sensor data, the processormay determine to activate one or more AI models. For example, the processormay detect a potential intruder based on the sensor data and activate an AI model. The processormay determine a countermeasure (e.g., a deterrence method) in response to the potential intruder. For example, the processormay generate a signal to activate the deterrence device.

The processormay be configured to generate a signal (e.g., INT). The processormay be configured to receive the signal IN, a signal (e.g., ACT) and/or a signal (e.g., RCD). The processormay be configured to communicate a signal (e.g., COM). The signal INT may comprise an intruder detection. The signal INT may be communicated to the memory. The signal ACT may comprise an action selection. The signal ACT may be received from the memory. The signal RCD may comprise a record data signal. The signal RCD may be received from the memory. The signal COM may comprise communication data. The signal COM may be communicated to and/or received from the communication device. The number, type and/or data communicated by the signals generated by and/or received by the processormay be varied according to the design criteria of a particular implementation.

The memorymay be configured to store data. In some embodiments, the memorymay be a local memory implemented by the apparatus. For example, the memorymay comprise a cache for the processor, a RAM for the processor, a storage device (e.g., hard disk drive, solid state drive, flash drive, etc.) for the apparatus, etc. In some embodiments, the memorymay be remote storage accessed by the apparatus. For example, the memorymay be remote storage device, a cloud storage/computing system, etc. In an example, the memorymay not necessarily be located in the environmentwith the apparatus. The type of the memoryimplemented may be varied according to the design criteria of a particular implementation.

The memorymay be implemented as part of a scalable computing service configured to store data, retrieve and transmit stored data, process data and/or communicate with other devices. The scalable computing service implementing the memorymay be implemented as part of a cloud computing platform (e.g., distributed computing). In an example, the scalable computing service implementing the memorymay be implemented as a group of cloud-based, scalable server computers. By implementing a number of scalable servers, additional resources (e.g., power, processing capability, memory, etc.) may be available to process and/or store variable amounts of data. For example, the scalable computing service implementing the memorymay be configured to scale (e.g., provision resources) based on demand. The scalable computing services implementing the memorymay implement scalable computing (e.g., cloud computing). The scalable computing may be available as a service to allow access to processing and/or storage resources without having to build infrastructure (e.g., the provider of the apparatusmay not have to build the infrastructure of the scalable computing service).

The memorymay be configured to receive the signal INT from the processor. The memorymay be configured to generate the signal ACT. The memorymay be configured to store information about various pests and/or particular deterrence actions available for the processor. The memorymay be configured to generate the signal RCD. The memorymay be configured to store event records generated by the processor.

The communication devicemay be configured to facilitate communication between the processor, one or more of the deterrence devicesand/or other networks. The communication devicemay implement a wired communication interface and/or a wireless communication interface. In some embodiments, the communication devicemay be part of the interface-. The type and/or protocol of communication implemented by the communication devicemay be varied according to the design criteria of a particular implementation.

The communication devicemay be configured to receive and/or generate the signal COM. The signal COM may comprise data communicated by the processorfor the deterrence deviceand/or to other devices/networks. The signal COM may comprise data communicated to the processorfrom the deterrence deviceand/or other devices/networks. The communication devicemay be configured to generate a signal (e.g., DT). The signal DT may be a countermeasure signal. The signal DT may be communicated to the deterrence device. The countermeasure signal DT may enable the deterrence deviceand/or provide instructions to the deterrence deviceto enable one or more available countermeasures implemented by the deterrence device.

The deterrence devicemay be configured to perform the countermeasure selected by the processor. The deterrence devicemay be configured to receive the signal DT. In some embodiments, the deterrence devicemay be configured to activate/deactivate in response to the signal DT. For example, the deterrence devicemay have a single function that may be turned on/off by the processor. In some embodiments, the deterrence devicemay be configured to execute computer readable instructions provided in the signal DT. The computer readable instructions may be generated by the processorto enable the deterrence deviceto perform particular actions to deter the pests.

The deterrence devicemay comprise one or more of drones, bots and/or fixed location devices. In an example, aerial drones may be configured to fly in the sky regionand/or maneuver to chase away the pests-and/or deploy countermeasures (e.g., spray water/chemicals, emit a scent, flash lights, etc.). Similarly, bots may comprise autonomous, ground-based devices that may be configured to chase away the pests-and/or deploy countermeasures (e.g., spray water/chemicals, emit a scent, flash lights, etc.). In an example, the bots may operate similar to a robotic vacuum cleaner and/or robotic lawnmower but be equipped with countermeasures. In some embodiments, the fixed location devices may be installed throughout the environment. For example, fixed location devices may comprise lights, speakers, fans, sprinkler systems, etc. The countermeasures may comprise a combination of lights, audio, chemicals, airflow, aggressive posturing, etc. Generally, the countermeasures implemented may be selected to be harmless and/or unnoticeable (or at least tolerable) to people. The countermeasures implemented by the deterrence devicemay be non-lethal. For example, the deterrence devicemay be implemented to remove the pests-from the environment, but not kill or otherwise cause harm to the pests-

The processormay comprise a block (or circuit), a block (or circuit)and/or a block (or circuit). The circuitmay implement an intruder module. The circuitmay implement a pest AI model. The circuitmay implement an efficacy AI model. The processormay comprise other components (not shown). The number, type and/or arrangement of the components of the processormay be varied according to the design criteria of a particular implementation.

The intruder modulemay be configured to detect an intruder in response to an analysis of the sensor data from the signal IN. The intruder modulemay be configured to implement an initial threshold detection. For example, the intruder modulemay provide sufficient analysis to detect that an event has occurred but may not necessarily perform a classification of the event. For example, the intruder modulemay provide sufficient analysis for detecting the presence of a potential intruder but may not necessarily determine whether the intruder is a person, an animal, a welcomed guest, a pest, etc. In one example, the threshold detection may comprise detecting a pre-determined amount of motion based on analysis of radar data. In another example, the threshold detection may comprise detecting particular types of objects based on performing computer vision operations. In yet another example, the threshold detection may comprise detecting particular temperature ranges in response to a temperature sensor and/or a thermal image. The type of initial threshold detection implemented may be varied according to the design criteria of a particular implementation.

The intruder modulemay be configured to generate a signal (e.g., EVT). The signal EVT may be an event signal. The intruder modulemay generate the signal EVT in response to an analysis of the sensor data that indicates an event has been detected (e.g., sensor data that crosses the initial threshold detection). The intruder modulemay be configured to activate one or more AI models implemented by the processor. In an example, the signal EVT may be presented to the pest AI model. For example, the pest AI modeland/or the efficacy AI modelmay operate in a low power, sleep (or standby) mode when not active. Operating in the low power mode may conserve computational resources and/or power consumption. The signal EVT may activate the pest AI modelto enable an analysis of the event detected by the intruder module.

The pest AI modeland/or the efficacy AI modelmay each implement a Large Language Model (LLM). In some embodiments, the processormay implement a single AI module configured to operate the functions of the pest AI modeland the efficacy AI model. In some embodiments, the processormay implement separate AI modules each configured to implement one of the pest AI modeland the efficacy AI model.

The pest AI modelmay be configured to generate the signal INT and receive the signal ACT and/or a signal (e.g., LRN). The signal INT may comprise a record of the event detected, the analysis of the event detected and/or the countermeasure selected in response to the event. The signal ACT may comprise information about known pests and/or available countermeasures that may be effective to deter the known pests. The signal LRN may comprise information about ongoing efficacy of particular deterrent methods for particular pests that may be used to update weights and/or biases that may define the pest AI model.

The pest AI modelmay be configured to perform a classification of the event. The pest AI modelmay be configured to analyze the sensor data received in the signal IN and/or the signal EVT. For example, the intruder modulemay be configured to extract from all of the available sensor data, the particular sensor data that may be relevant to the event detected by the intruder module(e.g., extract data based on timestamps that correspond to the detection of the event/intruder). The pest AI modelmay be configured to compare the sensor data in the signal EVT to a database of pests implemented by the memory. For example, the pest AI modelmay receive the signal ACT comprising information about various pests and/or available countermeasures for the pests. Based on the comparison, the pest AI modelmay be configured to perform a classification of the intruder. For example, the classification may determine whether the intruder is a pest at all and/or determine the particular pest detected. If the intruder is determined not to be a pest by the pest AI model, then the processormay perform no action/countermeasure (e.g., the pest AI modelmay return to a sleep/standby state). If the intruder is determined to be a pest by the pest AI model, then the pest AI modelmay determine which actions/countermeasures to perform to remove the pests-from the environment.

The pest AI modelmay be configured to determine a selection of a countermeasure in response to the classification of the intruder as a pest (e.g., the selected pest). The pest AI modelmay receive the signal ACT comprising the list of available countermeasures and/or data about methods of deterrence for the selected pest. The pest AI modelmay comprises a number of adjustable weights and/or biases that may be used to determine the appropriate countermeasure and/or a countermeasure most likely to be effective based on the context of the environment(e.g., time of day, number of people around, weather conditions, time of year, amount of noise, local knowledge based on past events specific to the environment, etc.) and/or the particular pests classification. The processormay generate the signal COM and/or enable the signal DT to activate and/or deploy the deterrence devicebased on a selected countermeasure determined by the pest AI model.

The pest AI modelmay continue to receive the sensor data generated by the sensors-while the deterrence deviceis deployed and/or after the deterrence devicehas finished performing the selected countermeasure. The pest AI modelmay analyze the environmentin order to monitor for an outcome of the selected countermeasure. For example, the pest AI modelmay determine whether the countermeasure has resulted in the pests-leaving the geo-fenced region, an amount of time taken to cause the pests-to leave the geo-fenced region, whether some, all or none of the pests-have left the geo-fenced region, whether the pests-have returned to the geo-fenced regionafter leaving, an amount of disturbance caused to people in the geo-fenced region, whether the pests-have been harmed, etc. Based on the outcome determined, the pest AI modelmay generate an event record.

The event record may comprise a text description of the classification determined, a reason for the classification of a particular pest, a countermeasure selected for the selected pest, a basis for the selection of the countermeasure, a description of the environmentand/or an outcome of the countermeasure. The signal INT may comprise the event record. The event record may be stored in the memory. For example, one event record may be generated for each event detected by the intruder module. The text description of the event record may comprise a natural text (e.g., human readable) description. The particular details generated for the event records may be varied according to the design criteria of a particular implementation.

The efficacy AI modelmay be configured to retrieve the event records from the memory. The efficacy AI modelmay be configured to request multiple event records from the memory. In one example, the efficacy AI modelmay periodically request event records of a particular type (e.g., based on the pest, based on a success/failure outcome, based on a time of day/year, based on the countermeasure selected, etc.). In another example, the efficacy AI modelmay request event records after each event record has been stored in the memory. The efficacy AI modelmay be configured to analyze multiple of the event records (e.g., pest events) stored in a database. The analysis may be configured to determine whether there is a decrease in efficacy of the countermeasure selection for a particular pest. The pests-may, over time, learn and/or adapt to one or more countermeasures. For example, the pests-may initially be scared away by a loud sound, but may eventually get used to the sound and no longer react, resulting in the sound countermeasure becoming ineffective. In another example, a countermeasure may be effective for large amounts of the pests but result in more disturbance to people in the environment, and at different times when there are fewer of the pests, a less intrusive countermeasure may be equally effective without disturbing the people (e.g., efficacy may be determined according to the reactions of the pests as well as an amount of disruption caused by the countermeasure).

The efficacy AI modelmay be configured to determine changes in efficacy of particular countermeasures for particular pests and/or environmental contexts. The efficacy AI modelmay be configured to determine errors in the classification of the pests made by the pest AI model. In one example, the efficacy AI modelmay determine a classification error (e.g., out of three goose events detected, one was actually a skunk). In another example, the efficacy AI modelmay determine a narrower classification (e.g., out of three goose events detected, one goose was a particular breed of geese that may have different behaviors). The efficacy AI modelmay be configured to adapt to a decrease in efficacy and/or effectiveness of countermeasures to the pests by modifying the countermeasure selection. In one example, the efficacy AI modelmay alter a countermeasure to attempt to improve efficacy (e.g., use a similar countermeasure with adjustments). For example, an initial countermeasure may be a sound generated with one frequency and/or sound signature and may be adjusted to be a sound generated with a different frequency and/or sound signature. In another example, the efficacy AI modelmay alter a countermeasure by selecting another type of countermeasure. For example, an initial countermeasure may be the sound generated, and an adjustment may be deploying one of the drones-

The efficacy AI modelmay determine updated parameters for the pest AI model. In an example, the updated parameters for the pest AI modelmay be adjusted weights and/or biases. The weights and/or biases may be used by the pest AI modelto determine which of the countermeasures to select and/or the classification of the pests. The updated parameters generated by the efficacy AI modelmay enable the pest AI modelto evolve over time. The efficacy AI modelmay generate the signal LRN comprising the updated parameters. The types of adjustments made by the efficacy AI modelmay be varied according to the design criteria of a particular implementation.

The weights and/or biases implemented by the pest AI modelmay be updated in response to receiving the signal LRN. The pest AI modelmay be configured to be activated in response to an event detected, by the intruder module, determine whether the event corresponds to a pest, classify the pests, determine which countermeasure to deploy, and generate an event record. The pest AI modelmay learn and/or adapt in response to the signal LRN. The pest AI modelmay continually react to events, and receive updates from the efficacy AI modelbased on the event records generated.